GOTO 2017 - Cloud Trends - Adrian Cockcroft
October 13, 2017
Overview
I enjoy watching Adrian Cockcroft speak. I think he has mostly useful things to say in a mostly practical and actionable way. Recently I viewed Adrian’s talk at the GOTO 2017 in Amsterdam. The entire talk is viewable on YouTube here: https://www.youtube.com/watch?v=EDZBYbEwhm8
In my opinion this is really two talks in one. The first talk is about company culture and how culture must be setup to enable innovation and scale. The second talk is that is that Function as a Service is the future and future cloud workloads will be dominated by training AI/Machine Learning for use at the device edge. There’s some AWS marketing sprinkled in as well.
Notes
Adrian starts talking about culture.
- Give people a yearning for the destination then let them get you there
- Give people a context for making good judgement NOT process
Netflix Culture
- Values are what we value - I guess this means the list is important?
- High Performance - You deliver or you are out
- Freedom and Responsibility - You are free to do what is needed but someone is watching. High Standard
- Context, not Control
- Highly Aligned, loosely coupled
- Pay top of the market
- Promotion and Development
NetFlix is extremely concentrated
- No REMOTE, two locations - engineering and movie production
- Netflix buys no-one
- Scales to a focused small company - couple of 100’s maybe a few 1000
Culture
- Intentional
- Appropriate
- Judgement - all cultures require this as the base
In 2008 there was a Netflix shock
- Two day outage due to F-ed SAN - IBM pSeries, Oracle & SAN
- Result: Move availability to application level
In 2009: Streaming began to replace DVD’s
Long talk on moving NetFlix to the cloud over years
The Future is monoliths -> micro-services -> functions
The future of the cloud is going to be for running AI/Deep Learning Training Model
The Challenge for AI Problems is SCALE
- PB of Data
- Training on 1000’s of GPU’s
Deploy Lamba Functions which have been Trained in the cloud to the edge
Amazon has created a Deep Learning AMI. Basically a Deep Learning Desktop. To hard to do yourself? Adrain seems to be over selling a bit here
Detail regarding Apache MXNet. Lots of reasons, including its not controlled by Google (like Tensorflow).
To Do Item: Follow up on DIY Robo Cars to learn this
Thoughts
- Continue education on ML and AI. Perhaps via a at home project like Adrian’s self driving toy car.
- Looks like AWS, Google and Microsoft are set to dominate/own the hardware for this market. Hopefully there’s space for more players then just these monsters. Having interacted with DigitalOcean I actually don’t think these companies are very good at Cloud UX.
- ML/AI market is fluid. Outside of a nascent self driving care market. Mostly what has been built are toys (Amazon Alexa) or Prototypes (Blue River Technology).